A new Paradigm of AI, Unsupervised In-Context Learning (ICL) of Large Language Models (LLM). Advanced In-Context Learning for new LLMs w/ 1 Mio token context length open up new possibilities for NEW Autonomous LEARNING of LLMs. Google DeepMind published new research on advanced ICL: ICL+. In this video we go from few-shot ICL (In-Context Learning) to many-shot ICL to improve the performance of our LLM (w/ 1M token) significantly. But note: it is domain specific and sensitive to causal reasoning patterns learned during pre-training of the LLM. A new Paradigm of AI, Unsupervised In-Context Learning (ICL). Forget Fine-Tuning: Unsupervised ICL Reinforced 00:00 1Mio token Context Length LLM ICL 05:26 From Few-Shot to MANY-Shot ICL 06:58 Performance jump 10:55 Benchmark data on ICL+ 16:05 Max Performance ICL 18:34 Reinforced and Unsupervised ICL 26:51 Autonomous Learning LLM by advanced ICL #airesearch #aieducation #ai